Finiteness and nonfinite forms

3. SYNTAX
Having defined nonfinite verb categories in terms of their functional differentiation from prototypical finite verbs, we must consider the syntax and semantics of nonfinite verbs not in absolute terms, but relative to the syntax and semantics of finite verbs. In other words, in what ways do nonfinite verb categories differ syntactically and semantically from prototypical finite verbs? In addressing the syntactic properties of nonfinite categories, I draw a broad distinction between internal and external syntax, that is, between the internal structure of phrases headed by nonfinite forms and the external syntactic properties of phrases headed by nonfinite forms, most importantly the syntactic functions of such phrases within superordinate clauses.

External Syntax

Prototypical finite verbs generally head clauses which either are unsubordinated (matrix) clauses or could stand as unsubordinated matrix clauses. In example 1a, the finite verb thought heads the matrix clause, while the finite verb knew heads the subordinate clause Bill knew Colin, which can function on its own as a matrix clause (example 1b):

As noted above, in some languages it is possible for some nonfinite verb categories to head some types of matrix clauses, but this is the more restricted situation, and by definition it cannot be the only possible use of a nonfinite verb form: The typical employment of a nonfinite verb is as the head of an embedded clause. The uses of nonfinite verbs correspond to the full range of ways in which clauses can be embedded. One use highly similar to that of finite verbs is as the head of a subordinate complement clause:¹²

In example 2, the infinitive (to) be heads the complement of the matrix verb seem, just as knew does in example 1a. The nonfiniteness of to be, in comparison with finite knew, is reflected in its internal syntax and semantics: The infinitival clause lacks an internal subject, at least on the surface, and lacks tense distinctions.
An even more prototypically verbal use of nonfinite forms is as part of a periphrastic expression, supplying the lexical content of the clausal predicate. In such constructions, as in subordinate complement clauses, nonfinite forms typically share their highest argument with the finite auxiliary, and lack certain morphosemantic properties, such as tense. Examples 3a, 3b, and 3c, respectively, illustrate the English gerund (eating), past participle (eaten), and bare infinitive (get) functioning as lexical heads of matrix clauses together with finite auxiliaries:⁴

In such constructions, we see nonfinite verbs coheading, alongside the finite auxiliary, clauses which are thoroughly verbal. Yet clauses headed by nonfinite verbs may also fill prototypically nonverbal roles in clauses, and it is here that the prototypical use of nonfinite forms is encountered.
Above, I draw a basic four-way distinction between verb, noun, adjective, and adverb as four major open lexical categories. Nonfinite forms whose nonverbal properties primarily correlate with the properties of nouns are often labeled gerunds(/action nominals/masdars), while those correlating with adjectives are participles, and those with adverbs, converbs.
The core use of gerunds is therefore to head clauses in fundamentally nominal functions, for instance, in argument positions:¹³

In example 4a, the gerund eating heads the phrase Bill eating ice cream, which serves as the subject of the matrix verb is. In example 4b, the gerund clause supplies the oblique agent of the passive verb, embedded under the preposition by. Similarly, in example 4c, the gerund clause functions as the object of the preposition without, the preposition phrase serving as an adjunct within the matrix clause. The internal syntax and semantics of such structures are discussed below, but note that in comparison with finite verbs, such embedded clauses may or may not have an explicit subject (examples 4a and 4b versus 4c), and do not distinguish tense.
The core use of participles is to head clauses with fundamentally adjectival functions, in particular, nominal attribution. In English, attributive adjectives occur prenominally, and prenominal modifiers must be final in their phrase, so this is restricted to intransitive or passive participles, as in examples 5a and 5b. There is no such restriction in German, as shown in example 6 (from Drijkoningen 1992, p. 55):


Adverbially used nonfinite clauses are typically headed by converbs, but it has long been known that participles and converbs tend to show overlapping functionality in many languages (I.V. Nedjalkov 1998, p. 451). Participles often adopt adverbial functions; V.P. Nedjalkov (1995) labels adverbially used participles “quasi-converbs,” while Haspelmath (1995, pp. 17–20) uses the term “copredicative participles.” The participles in example 7 exemplify such a construction; their use is adverbial in that they semantically modify the main predicate with an additional predication: eating ice cream in example 7a not only contributes additional information about Bill, but also contributes additional information about the process of his walking home:⁵

On the functionality of converbal participial clauses, see Haug (2008, 2010, 2012), Haug et al. (2012), and Lowe (2015).
One instance of a morphologically unambiguous participle restricted solely to adjectival use (i.e., attributive nominal modification) is the Hungarian present participle (de Groot 1995, pp. 287–88). Converbs restricted to adverbial use are far more common; the following example shows a Sanskrit converb (known as the absolutive) in typical function denoting a situation temporally prior to that of the main verb:⁶

Participles and converbs in many languages, including English, German, and Sanskrit, lack an explicit subject in their phrase; this gap is filled by an element of the superordinate clause; where participles show agreement, the element agreed with (i.e., the element modified by the participle clause) supplies the missing subject of the nonfinite clause. In some languages, the role of the missing argument is freer. In the following examples (Haspelmath 1994, p. 154), the Lezgian imperfective participle¹⁴